Concept
Augmented Synthetic Biology
The development cycle of synthetic biology yields faster and richer results when it’s enabled by Generative AI
It’s good to innovate – to create something that offers new alternatives. That’s what synthetic biology does. It involves redesigning organisms using engineering principles for transformative applications, resulting in novel products and processes, potentially at lower financial or environmental cost.
It’s used in areas including life sciences and healthcare; food and consumer products; and energy and sustainability. Challenges it typically faces include knowing and understanding metabolic pathways that could be engineered for improved yield; understanding the technical potential of the organization’s portfolio, and prioritizing molecules most likely to deliver results; and understanding the economic potential of that portfolio and prioritizing effort, weighing necessary investments against likely yields and current manufacturing processes.
Broader, deeper – and faster
But if you could find and explore more options – some of which may never have occurred to you – and you could do it faster, that, of course, would be better still.
Why? Because the traditional approach tends to be slow, expensive, and iterative. It starts with the identification of need – a business and technical assessment to establish goals. Organizations then undergo processes to identify new biological pathways, to propose new genes to code for the enzymes they need, to model the yield of those potential discoveries, and then to optimize their production. These phases typically entail a great deal of cyclical review before pathways can be deemed worthy of pursuit and taken to development at scale.
An approach powered by Generative AI (GenAI) and developed by Capgemini Engineering doesn’t necessarily do anything different. What it does do is to undertake each step faster, more inquisitively, and with a broader frame of reference than people can.
For instance, a pathway search (see first graphic) involves cataloguing potential models for enzyme development, and engineering the genetics of an organism to create new pathways. GenAI can employ a game-playing technique here, positing different options – some of them not humanly intuitive – and asking whether each of them in turn constitutes a good set of moves worthy of further exploration. What’s more, just as the AI can augment what people can do, so a human expert can guide the AI by setting parameters based on his or her domain knowledge.
GenAI can also help with predictive modeling. It can assess likely reactivity using ESM protein transformer and chemical transformer large language model (LLM) foundation models, for example predicting how an enzyme will catalyze a particular reaction for a particular molecule. This application is potentially groundbreaking: rates of reaction can vary greatly, and the approach developed here offers one of the most accurate levels of prediction that can be achieved in silico.
The Augmented Synthetic Biology also employs a database of curated enzyme catalyzed reactions for model training. However, this data is necessarily sparse. Using foundation LLMs on proteins and molecules separately allows the leveraging of complex structures in each from vastly bigger datasets. Bringing these models together with the sparse curated dataset achieves state-of-the-art accuracy in predicting reaction rates. This in turn is critical
to producing reliable results from the flux balance analysis modeling. All these elements help make better use of lab time, and so contribute significantly to the efficiency of the solution.
Rewarding inputs and positive outcomes at every stage
The initial business case, with its attendant considerations including cost and environmental impact, can be made solely by the organization. But at every subsequent stage of development (see second graphic), the GenAI-based approach Capgemini Engineering has developed can play a role.
Design, build, test, and learn processes are still cyclical, but they are augmented. It’s a solution that researches more widely, delves more deeply, cross-checks more extensively, and tests with what-if scenarios more rigorously. It doesn’t provide answers, but it suggests promising possibilities at every step, narrowing the focus, increasing the prospects of a viable solution, and saving considerably on budget – because the further right you go in the pipeline shown, the more you move into in-vitro and in-vivo, and the more expensive each project becomes.
Our AI Toolkit can be integrated into a bio-assessment pipeline:
Potential applications are wide-ranging. Capgemini Engineering’s GenAI-based offer enables organizations to explore new approaches in waste-water treatment, in the development of aromas and flavors in food processing, in the creation of new therapeutic pharmaceuticals, in the development of more effective chemical processes, and in effecting changes in the behavior of chemical products in given circumstances.
The word ‘enabling’ here is key: the generative AI doesn’t itself make decisions or create new organisms, so there are no regulatory or ethical concerns. Rather, in highlighting promising pathways, it’s accelerating and enhancing regular human scientific endeavor.
This is an area that will have a major impact on people’s lives. Small wonder, then, that over the next 10-20 years, it’s estimated that the annual economic impact of synthetic biology will be $2-4 trillion.[1]
These are exciting times.
Augmented Synthetic Biology – what’s involved:
Pathway searching – game-playing models predict promising avenues of research
Increased efficiency – research focused and fine-tuned by GenAI reduces costs
Faster, more promising outcomes – GenAI has applications at every stage
[1] * McKinsey “The Bio Revolution: Innovations transforming economies, societies, and our lives.”
